You are testing a new trade method with other people's real money? Why don't you run a backtest or WFA before?
Trading algorithmically a folio without stops (with IB), real $$$
- Thread starter fullautotrading
- Start date
If "investors" are too stupid to let this happen, why shouldn't he?
he can also test his psychology on live money and when the money is lost
and IF nobody says anything, he can always move to other investors who
are eagger to lose their money.
A quick mid week update. Metals did not move, so we need to wait for them to move and recover
Today, there was some activity with miners and energy (ERX) moving a bit, especially towards the end of the session. Mkt seems to be heading down.
USLV has gone a little too far for my taste, but it will be recovered. The reason why these metal ETFs have been loading a bit too much is due essentially to the fact that I have been using the "Bias" game, with a "short only position constraint".
While this setting seemed to work fine with the (inverse) 2X, with the 3X it seems to absorb too much "investment" due to the 3X amplification of the price waves. And the fact that we had too many metals layers with all those 3X has made this particularly evident. (Well, of course it's all relative to the available capital.)
To control this aspect I have added a new flag to the "game rules", which allows overriding the instrument "position constraint", for the players open in "hedging-protection" mode. This way, one can still use the bias game with the short bias, but also allow for a smooth and gradual position "inversion" if hedging is needed.
Clearly, now having to hedge a bit too "late" (relative to capital), makes the procedure a bit clumsy because the hedging orders will need to be larger, and so the possible reversing entries (in case the instrument then reverses).
To automatize this aspect, I have been introducing a new sizing mode (a new rule of the game), which ensures that a given fraction of the opposite side players will be hedged by each new order. This would allow a smooth continuous hedging, without having to do late clumsy hedging actions.
Working on the edge of the available funds has also suggested another interesting improvement to deal with order rejections due to insufficient margins. Previously, I had already addressed this issue, when we had the nasty UGAZ mishap. My previous solution, in response to the problem, was to automatically appropriately reduce the orders size in order to allow filling (otherwise the instrument remains practically "locked", and can have large drawdown).
Now I have found a good refinement, which was actually suggested by the actions carried out manually to deal with insufficient margins. While the above mechanism remains in place, I have added the fact that when there are order rejections due to insufficient funds, only one player at a time is closed, and it's selected to be the closest to the current price (bid/ask, depending if it's a buy/sell player).
The reason why that is useful is that closing the nearest players, the new funds that are freed when the price moves can be possibly used to allow the farthest players to close. Thus minimizing the disruption of the algorithm due to the order rejections.
Today, there was some activity with miners and energy (ERX) moving a bit, especially towards the end of the session. Mkt seems to be heading down.
USLV has gone a little too far for my taste, but it will be recovered. The reason why these metal ETFs have been loading a bit too much is due essentially to the fact that I have been using the "Bias" game, with a "short only position constraint".
While this setting seemed to work fine with the (inverse) 2X, with the 3X it seems to absorb too much "investment" due to the 3X amplification of the price waves. And the fact that we had too many metals layers with all those 3X has made this particularly evident. (Well, of course it's all relative to the available capital.)
To control this aspect I have added a new flag to the "game rules", which allows overriding the instrument "position constraint", for the players open in "hedging-protection" mode. This way, one can still use the bias game with the short bias, but also allow for a smooth and gradual position "inversion" if hedging is needed.
Clearly, now having to hedge a bit too "late" (relative to capital), makes the procedure a bit clumsy because the hedging orders will need to be larger, and so the possible reversing entries (in case the instrument then reverses).
To automatize this aspect, I have been introducing a new sizing mode (a new rule of the game), which ensures that a given fraction of the opposite side players will be hedged by each new order. This would allow a smooth continuous hedging, without having to do late clumsy hedging actions.
Working on the edge of the available funds has also suggested another interesting improvement to deal with order rejections due to insufficient margins. Previously, I had already addressed this issue, when we had the nasty UGAZ mishap. My previous solution, in response to the problem, was to automatically appropriately reduce the orders size in order to allow filling (otherwise the instrument remains practically "locked", and can have large drawdown).
Now I have found a good refinement, which was actually suggested by the actions carried out manually to deal with insufficient margins. While the above mechanism remains in place, I have added the fact that when there are order rejections due to insufficient funds, only one player at a time is closed, and it's selected to be the closest to the current price (bid/ask, depending if it's a buy/sell player).
The reason why that is useful is that closing the nearest players, the new funds that are freed when the price moves can be possibly used to allow the farthest players to close. Thus minimizing the disruption of the algorithm due to the order rejections.
I do extensive simulations, however the real world always offers challenges that a simulation environment may not fully incorporate and that may prove to be crucial in the grand scheme of things (execution problems, margin impact and issues, leverage, spread issues, shortability issues, decay, contango, drifts, delivery issues, dividends, interests, ... etc.). I always carry small tests in my own accounts before each update, but very good capital is necessary for meaningful real world applications and real advances. The fact that we have been executing thousands and thousands of limit orders on a large number of instruments, keeping track of all complete historical charts and of the details of all the single executions and thousands of logical players, through systems running from 2 different continents and the market on the other side of the ocean, without the smallest glitch, and running continuously 24/7 should provide some indication of that. (If you compare that with the need of the TWS itself to shut down every day, that already may tell you something.)
Remember that things are not born perfect. It's a continuous process of growth and refinement. After all, the first flight lasted 12 seconds and covered 120 feets. (And, I guess that at the time the Wright brothers also had their denigrators.) For this reason I am mostly grateful to the investors who, in addition to trading their funds, also make available some resources and support to expand this research and journey.
As to "backtest" they may have no (statistical) relation at all with statistical properties of a trading procedure. It is, in fact, even possible in principle (but also in reality) to have a trading methodology with good statistical properties which on a past realization is showing performances falling on the extreme left tail.
On the other hand, you may have (actually almost always) a trading "strategy" which shows good "performances" on that only past realization, but which statistically or logically makes no sense at all. In fact, if you take and place them in my simulator, you see in a blink of an eye they have no statistical ground (and often no logical meaning).
Actually, this is the by far most common case, as in fact most of the beginners with no statistics background (or simply not actually "thinking" yet) make the common logical inversion to select trading strategies and adjust (through iteration on the parameter space, instead of the price-curve space) parameters in such a way to have (conditional) good results on that past realization. Which, clearly, is a formal fallacy.
It depends. A simple backtest can not verify, but it can falsify a system. Complex backtests with WFA and Monte Carlo analysis can give you some estimate with which you then can compare the live performance of your system. This allows you to pull out early when the system is out of the borders determined by the test.
That's why I was asking you about the comparison of your live trading with a simulation. Your system is running since 8 months and has not produced any profit. This would normally be a reason to abandon it. I run many automated systems, but I would not continue a system that does not show a clearly rising equity curve at least after a few months.
But of course such a decision depends on what you expect from the system, and such expectation is normally based on backtests and simulations.
That's why I was asking you about the comparison of your live trading with a simulation. Your system is running since 8 months and has not produced any profit. This would normally be a reason to abandon it. I run many automated systems, but I would not continue a system that does not show a clearly rising equity curve at least after a few months.
But of course such a decision depends on what you expect from the system, and such expectation is normally based on backtests and simulations.
>It depends. A simple backtest can not verify, but it can falsify a system.
That statement is obviously false. As a single realization has probability 0.
>Your system is running since 8 months and has not produced any profit.
That statement is false. As we are on (solar) day # 149 of the current trading session (including holidays and weekends).
Also, you show no familiarity with real world equity curves. A real world curve will normally show "load" and "discharge" phases. Those equity curves you see going straight up are just some interpolations of the past ("curve fitted delusions"), done by computer.
Later, I will make a post on the logical necessity of the DD for consistent profitability.
The actual important practical point is tuning the hedging mechanism, to keep the DD within the boundaries one likes (clearly, relative to the available resources). And that is a task which may also depend on the structural characteristics of a single instrument. (To me seeing the PNL fluctuate at about -10%-15% of the available capital does not sound too uncommon. But of course, if you can do better, you can always point us to your journal.)
>This allows you to pull out early when the system is out of the borders determined by the test.
This is something I guess that you invented on your own, or read somewhere. In my approach there no such a thing as "pulling out" (unless for technical or other unpredicted issues). It's a continuous and uninterrupted process of scalping and hedging.
That statement is obviously false. As a single realization has probability 0.
>Your system is running since 8 months and has not produced any profit.
That statement is false. As we are on (solar) day # 149 of the current trading session (including holidays and weekends).
Also, you show no familiarity with real world equity curves. A real world curve will normally show "load" and "discharge" phases. Those equity curves you see going straight up are just some interpolations of the past ("curve fitted delusions"), done by computer.
Later, I will make a post on the logical necessity of the DD for consistent profitability.
The actual important practical point is tuning the hedging mechanism, to keep the DD within the boundaries one likes (clearly, relative to the available resources). And that is a task which may also depend on the structural characteristics of a single instrument. (To me seeing the PNL fluctuate at about -10%-15% of the available capital does not sound too uncommon. But of course, if you can do better, you can always point us to your journal.)
>This allows you to pull out early when the system is out of the borders determined by the test.
This is something I guess that you invented on your own, or read somewhere. In my approach there no such a thing as "pulling out" (unless for technical or other unpredicted issues). It's a continuous and uninterrupted process of scalping and hedging.
The first post of this thread was in October 2013, and we have now June 2014 and $30K loss, so to me that looks like 8 months with no profit, even with benevolent consideration.
Your permanent remarks like "obviously false" and "curve fitted delusion" are no arguments. If something is obviously false only to you, but to no one else, then it's obviously not so obvious. Such statements with no further explanation just appear as arrogance and lack of knowledge.
There are no such things as "load" and "discharge" in an equity curve. This is marketing gibberish. Live trading an automated system normally produces an equity curve like this:
This is the first 8 months of one of our systems, on a Zulutrade signal account. You can see the drawdowns, but you can also see a clear tendency. The PnL curve of your system has no tendency, or if there is one, it is downwards. Neither me nor anyone else here has understood how your system is supposed to produce profit. You have never backtested it. Yet you're trading it live here with other people's money, and intend to continue that without validating if the basic idea, whatever it might be, works at all. That's what I am not understanding.
Your permanent remarks like "obviously false" and "curve fitted delusion" are no arguments. If something is obviously false only to you, but to no one else, then it's obviously not so obvious. Such statements with no further explanation just appear as arrogance and lack of knowledge.
There are no such things as "load" and "discharge" in an equity curve. This is marketing gibberish. Live trading an automated system normally produces an equity curve like this:
This is the first 8 months of one of our systems, on a Zulutrade signal account. You can see the drawdowns, but you can also see a clear tendency. The PnL curve of your system has no tendency, or if there is one, it is downwards. Neither me nor anyone else here has understood how your system is supposed to produce profit. You have never backtested it. Yet you're trading it live here with other people's money, and intend to continue that without validating if the basic idea, whatever it might be, works at all. That's what I am not understanding.
>The first post of this thread was in October 2013, and we have now June 2014 and $30K loss, so to me that looks like 8 months with no profit, even with benevolent consideration.
In addition to problems with logic and basic statistics, you also appear to have problems with reading.
I am not sure why you bother commenting without even having read the thread.
Frankly, your contribution to this thread has been nothing so far, just an unhelpful continuous attempt to discredit. You are not interested in ideas, but only in a scarcely mature attitude of arguing from ignorance.
I must start suspecting you are one of those individuals who prefer trolling, instead of devoting their life to meaningful scientific construction. Hope to see more meaningful comments from you in the future.
>This is the first 8 months of one of our systems, on a Zulutrade signal account. You can ... see a clear tendency.
"Signals" ?. Ah, now it's finally pretty clear. I do see a tendency
Remember the "laws of attraction". The actions you do, will determine mostly the kind and the level of people who you will find around you.
In addition to problems with logic and basic statistics, you also appear to have problems with reading.
I am not sure why you bother commenting without even having read the thread.
Frankly, your contribution to this thread has been nothing so far, just an unhelpful continuous attempt to discredit. You are not interested in ideas, but only in a scarcely mature attitude of arguing from ignorance.
I must start suspecting you are one of those individuals who prefer trolling, instead of devoting their life to meaningful scientific construction. Hope to see more meaningful comments from you in the future.
>This is the first 8 months of one of our systems, on a Zulutrade signal account. You can ... see a clear tendency.
"Signals" ?. Ah, now it's finally pretty clear. I do see a tendency
Remember the "laws of attraction". The actions you do, will determine mostly the kind and the level of people who you will find around you.
Let's focus on some more interesting topic. I promised earlier to shed some light on the mechanisms to use instrument correlations to improve hedging across layers.
For instance, if we are trading TNA and TZA, since the general behavior of one of the two is pretty much determined by the other one, it's clear that it can be useful for an automated trading algo to balance the exposures of correlated assets.
To this purpose, I have defined for each Instrument, say I, the following quantities:
SAE(I) := "Signed (Volatility-Adjusted) Exposure"
REC(I) := "Relative Exposure of Correlated Instruments"
Such quantities are defined as follows:
SAE(I) := Closing Price * Position with sign * Multiplier * Volatility * 0.001 (so, say with $K as unit of measure)
[clearly, may multiply by 0.00001, if annualized volatility is in %]
for instance, an instrument could have an exposure of 15.56K, and so on.
While, the "Relative Exposure of Correlated Instruments" is defined as:
Where corr(I,J) indicates your preferred correlation (or "codirection") metrics (such as for instance the SCX, we have seen previously).
This quantity, along with the instrument exposure SEA(I) is useful to rebalance exposure across layers, for instance, according to the following scheme:
The meaning of the schema seems intuitive.
Let's consider, for instance, SAE(I) > 0 and REC(I) > 0. In this case, both the instrument I and the set of correlated instruments C(I) have a positive exposure. So a SELL order of I would have an "hedging" effect on the exposure.
Let's consider for instance SAE(I) < 0 and REC(I) > 0. In this case, I has "negative" exposure (it has a short position), while the correlated instruments are "contrasting" this exposure (either through opposite position under positive correlation or same side position under negative correlation). Now, depending on which of the 2 exposure is larger (REC(I) or SAE(I)), we have an exposure reduction either by Selling of Buying (respectively) I. The other 2 cases are analogous.
Similarly, they offer an indication of the "role" of I within the folio. For instance, if SAE(I) > 0 and REC(I) < 0 and |SAE(I)| < |REC(I)| we can interpret the instrument I as having an "hedging value" within the folio and that it's suitable for (signed) position increase. The interpretations of the other cases, can be done applying a similar logic.
Let me know if you see improvements or errors, so I can fix them.
For instance, if we are trading TNA and TZA, since the general behavior of one of the two is pretty much determined by the other one, it's clear that it can be useful for an automated trading algo to balance the exposures of correlated assets.
To this purpose, I have defined for each Instrument, say I, the following quantities:
SAE(I) := "Signed (Volatility-Adjusted) Exposure"
REC(I) := "Relative Exposure of Correlated Instruments"
Such quantities are defined as follows:
SAE(I) := Closing Price * Position with sign * Multiplier * Volatility * 0.001 (so, say with $K as unit of measure)
[clearly, may multiply by 0.00001, if annualized volatility is in %]
for instance, an instrument could have an exposure of 15.56K, and so on.
While, the "Relative Exposure of Correlated Instruments" is defined as:
Where corr(I,J) indicates your preferred correlation (or "codirection") metrics (such as for instance the SCX, we have seen previously).
This quantity, along with the instrument exposure SEA(I) is useful to rebalance exposure across layers, for instance, according to the following scheme:
The meaning of the schema seems intuitive.
Let's consider, for instance, SAE(I) > 0 and REC(I) > 0. In this case, both the instrument I and the set of correlated instruments C(I) have a positive exposure. So a SELL order of I would have an "hedging" effect on the exposure.
Let's consider for instance SAE(I) < 0 and REC(I) > 0. In this case, I has "negative" exposure (it has a short position), while the correlated instruments are "contrasting" this exposure (either through opposite position under positive correlation or same side position under negative correlation). Now, depending on which of the 2 exposure is larger (REC(I) or SAE(I)), we have an exposure reduction either by Selling of Buying (respectively) I. The other 2 cases are analogous.
Similarly, they offer an indication of the "role" of I within the folio. For instance, if SAE(I) > 0 and REC(I) < 0 and |SAE(I)| < |REC(I)| we can interpret the instrument I as having an "hedging value" within the folio and that it's suitable for (signed) position increase. The interpretations of the other cases, can be done applying a similar logic.
Let me know if you see improvements or errors, so I can fix them.